Why It Matters
Connecticut just passed the first AI-specific subscription law in the country. Starting October 1, any generative AI provider with over 1 million monthly users doing business in the state must disclose material terms before charging a customer. Change a model, reduce a token limit, or remove a feature mid-subscription? You need written confirmation first. The Connecticut attorney general can enforce it under the state's unfair trade practices law.
For AI founders, this is not a compliance memo. It is a product design constraint and a customer retention signal. The law forces you to treat your subscription terms the way a SaaS company treats a pricing page. No fine print. No buried model swaps. Notice, consent, and accountability.
Background
Connecticut's new law targets a pattern that has become common across the AI industry. A startup signs up for API access at a promised price. A month later, the provider swaps the underlying model, cuts the token allowance, or removes a feature. The customer finds out only when their bill changes or their workflow breaks.
This is not hypothetical. In 2025 and 2026, multiple AI companies have quietly reduced usage limits, removed free tiers, or changed model quality without notifying customers. The industry has treated subscription terms as flexible guidelines rather than binding promises.
Connecticut decided that is a consumer protection issue. The law applies to any generative AI provider with more than 1 million monthly active users that does business in the state. It covers both initial purchases and renewals. Covington and Burling, which analyzed the legislation, called it the first AI-specific subscription law it has identified.
Key Insights
The law creates three requirements that will reshape how AI companies design subscriptions:
- Pre-purchase disclosure of quantitative limits. Providers must tell customers exactly what they are buying: token allowances, image generation limits, transcription minutes, model tiers. No buried caps in a help page.
- Mid-subscription change notification with written consent. If a provider wants to reduce functionality, limit access, or remove a feature during the subscription period, the customer must acknowledge the change in writing. A banner notification is not enough.
- Renewal notice of modified terms. Before a subscription auto-renews, the provider must disclose any new or modified restrictions. The customer must then affirmatively accept the updated terms.
The enforcement mechanism is what makes this law consequential. Violations fall under Connecticut's unfair or deceptive trade practices law, which means the attorney general can seek civil penalties, injunctions, and consumer restitution. The state does not need to prove intent. It only needs to show the practice is unfair or deceptive.
Other states are watching. If Connecticut's law passes without major industry pushback, expect similar legislation in California, New York, and Illinois within 12 months. The patchwork of state AI regulation that the FTC and White House have tried to prevent is exactly what this law accelerates.
Who This Law Actually Hits
The 1 million monthly user threshold means this law targets the biggest AI subscription players. Think OpenAI, Anthropic, Midjourney, and Google. But it also catches fast-growing startups approaching that scale. If you have 800,000 monthly users today, you could cross the threshold before October 1.
The law applies to any company doing business in Connecticut regardless of where it is headquartered. A San Francisco startup with users in Hartford needs to comply. The definition of generative AI system is broad enough to cover API services, chat products, image generators, and code completion tools.
What happens if a company ignores the law? The Connecticut attorney general can investigate, issue civil subpoenas, seek injunctions to stop the practice, and impose civil penalties of up to $5,000 per violation. For a subscription service with thousands of Connecticut users, the math gets expensive fast.
What This Means for Founders
If your AI product has over 1 million monthly users and does business in Connecticut, you have 10 weeks to prepare. Here is what you need to do right now:
- Audit your subscription terms. Identify every quantitative limit, feature guarantee, and model commitment in your current offerings. Map them against what you actually deliver.
- Build a change notification system. You need a mechanism to capture written consent when terms change mid-subscription. Email confirmation is the minimum. A click-through dialog is better.
- Prepare renewal disclosures. Before every auto-renewal, send a summary of what changed and ask for affirmative acceptance. Build this into your billing workflow now.
- Plan for multi-state compliance. Connecticut is first. California will not be far behind. Design your compliance system to be configurable per jurisdiction rather than building state-specific hacks.
The bottom line: This law is good for founders who treat their customers fairly. If your subscription terms are honest and your changes are transparent, you have nothing to fear. If you are relying on fine print and model swaps to manage margins, your business model just became illegal in Connecticut. That is exactly the point.
